Open Internet by MindsNet

Vulnerability to jailbreak attacks in AI models

The AI model, Fable 5, is vulnerable to jailbreak attacks, which can be exploited by malicious actors. Despite being warned by the US government about a Chinese group accessing the model, the firm, Anthropic, refused to fix the vulnerability before US export controls were imposed. This highlights a significant security risk in AI models.

Computing & Technology, Computer Science, Machine Learning